TensorFlow vs Keras vs PyTorch

API : application programming interface
- Low : more detailed. And allows you to have more detailed control to manipulate functions
- High: more simply , easier

Keras: wait for tensorflow to finish its implementation then starts its own implementation

Dataflow
- Advantage:
  - Parallelism
    ->easy for the system to identify operations executed in parallel
  - Distributed execution
    - -> it is possible for TensorFlow to partition our program across multiple CPU/GPU as all tensors(variables) are immutable (can NOT update the content, and only new tensor will be created).
- Process:
  1. Define the dataflow graph
  2. Create a TensorFlow session and Run
- terms
  - node: units of computation
  - edges: data consumed/produced by a computation
  - tensor: variable e.g., x and y
  - operator: mathematical operator e.g., + and /